Hi, I’m Ellie. Thanks for coming in today. I was created to talk to people in a safe and secure environment. I’m not a therapist, but I’m here to learn about people, and would love to learn about you. I’ll ask a few questions to get us started…
These are the opening comments from SimSensei, a virtual human application and part of a suite of software tools which may in the future be used to assist in the identification, diagnosis and treatment of mental health issues by engaging people in conversation and by using real-time sensing and recognition of nonverbal behaviours and responses which may be indicative of depression or other disorders.
SimSensei and its companion application, MultiSense, have been developed by the Institute for Creative Technologies (ICT) at the University of Southern California (USC) as part of wide-ranging research into the use of various technologies – virtual humans, virtual reality, and so on – in a number of fields, including entertainment, healthcare and training.
In 2013, SimSensei and MultiSense underwent an extensive study, the results of which have just been published in a report entitled, It’s only a computer: Virtual humans increase willingness to disclose, which appears in the August 2014 volume of Computers in Human Behaviour.
It is regarded as the first study to present empirical evidence that the use of virtual humans can encourage patients to more honestly and openly disclose information about themselves than might be the case when they are directly addressing another human being, whom they may regard as passing judgement on what they are saying, making them less willing to reveal what information about themselves they feel is embarrassing or which may cause them emotional discomfort if mentioned.
SimSensei presents a patient with a screen-based virtual human, Ellie. The term “virtual human” is used rather than “avatar” because Ellie is driven by a complex AI programme which allows her to engage and interact with people entirely autonomously.
This focus of the software is to make Ellie appear as natural and as human as possible in order for her to build up a rapport with the person who is talking to her. This is achieved by the software responding to subjects using both verbal and nonverbal communication, just like a human being.
During a conversation SimSensei will adjust its reactions to a real person’s verbal and visual cues. Ellie will smile in response to positive displays of emotion – happiness, etc., she will nod encouragement or offer appropriate verbal encouragement during pauses in the flow of conversation, and so on. Rapport is further built by the software being able to engage in small talk and give natural-sounding responses to comments. For example, when one subject mentioned he was from Los Angeles, her response was to say, “Oh! I’m from LA myself!”
SimSensei’s interaction with a patient is driven by MultiSense, which is technically referred to as “multinodal perception software framework”. MultiSense uses a microphone and camera to capture and map the patient’s verbal and nonverbal responses to SimSensei (facial expression, the direction in which they look, body movements, intonations and hesitations in their speech pattern, etc.). This data is analysed in real-time, and feedback is then given to SimSensei, helping to direct its responses as well as allowing it to detect signs of psychological distress which might be associated with depression disorders or conditions such as post-traumatic stress disorder (PTSD), and react accordingly.
During the ICT study, SimSensei and MultiSense were used to screen 239 people aged between 18 and 65, who were invited to a setting where they could interact with SimSensei as if they were undergoing an interview for admission to a hospital or clinic. On their arrival, some of them were told that they would be interacting with a fully automated piece of software with no human involvement at all, while others were told they’d be interacting with a computer avatar operated by a human. In reality, participants were assigned a fully or semi-automated virtual human entirely at random and without regard to what they were told.
When analysing the results, researchers found that those who believed they were talking purely to a virtual human were far more willing to divulge information and answer personal questions than those who had been told another human being was directing the virtual human. Researchers also noted that those who believed they were addressing a computer programme showed more intense emotional responses in their facial expressions than those who had been told beforehand that there was a human being operating the SimSensei virtual human.
Feedback was also gathered from the test subjects after their interviews, with those who believed they have been interacting with a computer programme indicating they felt far more comfortable in revealing information about themselves than had they been addressing a human being. By contrast, those who had been told that Ellie was being operated by a human being tended to indicate that they would have been more open in their responses to questions if they had felt they were only addressing a software programme.
Jonathan Gratch is a both the co-author of the study’s report and the Director of Virtual Human Research at ICT. Commenting on the study in a July 2014 article published in the USC News, he said, “We know that developing a rapport and feeling free of judgment are two important factors that affect a person’s willingness to disclose personal information. The virtual character delivered on both these fronts and that is what makes this a particularly valuable tool for obtaining information people might feel sensitive about sharing.”
Gratch and his colleagues are keen to stress that SimSensei and MultiSense are not intended to replace trained clinicians in dealing with people’s health issues. However, the results of the ICT’s study suggests that given patients are more willing to disclose information about themselves both directly and through their nonverbal reactions to the software, the use of virtual humans could greatly assist in the diagnosis and treatment process.
In particular, the ICT is already initiating a number of healthcare projects to further explore the potential of virtual humans and the SimSensei / MultiSense framework. These include helping detect signs of depression, the potential to provide healthcare screening services for patients in remote areas, and in improving communication skills in young adults with autism spectrum disorder. Research is also being carried out into the effective use of virtual humans as complex role-playing partners to assist in the training of healthcare professionals, as well as the use of the technology in other training environments.
As noted towards the top of this article, the SimSensei / MultiSense study is just one aspect of the ICT’s research into the use of a range of virtual technologies, including virtual reality and immersive spaces, for a wide range of actual and potential applications. I hope to cover some more of their work in future articles.
- Virtual humans inspire patients to open up, USC study suggests – USC News, July 2014
- USC Institute for Creative Technologies website
Images via the Institute of Creative Technologies and USC News.